Babylab data - exploration

Since last time:

To do:

1 Intro

1.1 Context

One of the goals of sepages is to analyse the link between exposure to endocrine disruptors during pregnancy and the infant’s neurological development. One of the experiments to quantify neurologic development was to perform eye tracking experiments at Grenoble’s LPNC BabyLab. Sepages infants were subject to several tasks and were seen up to 3 times: at 5 months, 12 months, 24 months.

Here are the descriptive stats on the babylab dataset DataFile_030220_anonym_id.csv.

First we present the different tasks and scores. Then we briefly describe the population. Finally we decide on a subselection of scores for final analysis.

1.2 Eye tracking

Eye movement is measured by X-Y coordinates continuously. Eye movement consists of a sequence of saccades (movement) and fixations (non movements). Using the eye tracker several tasks are performed for which various scores were computed. Four tasks were performed for this study:

  • scene perception
  • face perception and recognition
  • saccades to target
  • smooth pursuit

1.2.1 Task 1: Scene perception (#scene)

An image is shown to the baby for five seconds each. The experiment is repeated six times.

We are interested in looking at what part of the image the infant looks at.

The scores available for this task are

  1. The number of saccades during the task (sacc_n_sce)
  2. The values of a and b in the equation v = b*d^a (p1_sce & p2_sce)
  3. The mean number of fixations over the six images (fix_n_sce)
  4. The mean total time spent looking at the image (look_t_sce)
  5. The mean duration of one fixation (fix_dur_sce)
  6. A score to compute if the child looks at the same place of the image or not as the other children (map_sce)

1.2.2 Task 2: Face perception and recognition (#face)

Two pictures of the same human face are simultaneously shown to the baby for 5 seconds. Then, one picture is replaced by a new one (the faces have different levels of attractiveness) and the two different pictures are shown to the baby for 5 seconds. The experiment is repeated four times.

We are interested in measuring the reaction to novelty: the baby should preferentially look at the new face.

The scores available for this task are

  1. The number of saccades during the task (sacc_n_fac)
  2. The values of a and b in the equation v = b*d^a (p1_fac & p2_fac)
  3. The mean number of fixations over the six images (fix_n_fac)
  4. The mean total time spent looking at the image (look_t_fac)
  5. The mean duration of one fixation (fix_dur_fac)
  6. The reaction to novel face over four tests (novelty_fac) need more info on formula
  7. Reaction time (react_t_fac) need more info on formula
  8. Percent time spent looking at eyes (pct_eyes_fac)
  9. Percent time spent looking at mouth (pct_mouth_fac)

1.2.3 Task 3: Saccades to target (#target)

Attention of child is fixed at the center of the screen with a target. When the attention of the child is fixed on the target in the middle of the screen, the target disappears and re-appears in one of the 8 peripheral positions. The child then saccades toward the new target until it fixes it. The target then switches to the middle and the experiences is repeated 8 times.

The scores available for this task are

  1. Reaction time (reac_t_tar) need more info on formula
  2. Mean length of the first saccade to target (dist_tar)

1.2.4 Task 4: Smooth pursuit (#pursuit)

There is a central fixation point. Once the point disappears a new point appears on the right and starts to move. The child then pursues the target that does three circles (not sure of the N). Then the target reappears in the middle and the experiment is repeated 4 times.

[ADD IAMGE]

The scores available for this task are

  1. Reaction time (reac_t_pur) need more info on formula

2 Population description

2.1 Individuals

  • N unique individuals = 188
  • Sex:
n % val%
F 83 44.1 44.1
M 105 55.9 55.9

As a reminder in sepages: 251 boys (53%) and 218 girls (47%).

  • Birth dates:

2.2 Experiments

  • N experiments: 264

Task 2 missing for first 5 (not yet done) and when all signal was bad. (cf. notes meeting with DM 11/03/2020). Same for task 4. These tasks are more “difficult” so it’s harder to get valid signal.

2.3 Age at experiment

Older outlier:

## # A tibble: 1 x 4
##   date_birth date_exp   age_cat age_days
##   <date>     <date>       <dbl>    <dbl>
## 1 2016-12-07 2008-01-19      24     1097

True age in months = 1097 / 30.4 = 36…

Error? Exclude?

Three groups of age: 5 months, 12 months and 24 months with a majority of 24 months. Categorical age var:

n % val%
5 46 17.4 17.4
12 67 25.4 25.4
24 151 57.2 57.2

2.4 Repeated measures

Number of repeated data:

  • 118 children with 1 experiment
  • 64 children with 2 experiments
  • 6 children with 3 experiments

2.5 N with exposure

2.5.1 eye tracker 5 months

period n
exposure_T1 46/46 (100%)
exposure_T3 46/46 (100%)
exposure_M2 20/46 (43%)
exposure_Y1 14/46 (30%)

2.5.2 eye tracker 12 months

period n
exposure_T1 67/67 (100%)
exposure_T3 67/67 (100%)
exposure_M2 14/67 (21%)
exposure_Y1 13/67 (19%)

2.5.3 eye tracker 24 months

period n
exposure_T1 151/151 (100%)
exposure_T3 150/151 (99%)
exposure_M2 56/151 (37%)
exposure_Y1 40/151 (26%)

3 Eye tracker scores

First we do some univariate statistics on the scores then we look at relations between them to see if we can/need to exclude some from further analysis.

3.1 Description

Data Frame Summary

scores
Dimensions: 264 x 28
Duplicates: 0
No Variable Label Stats / Values Freqs (% of Valid) Graph Missing
1 ident [character] Unique child id 1. 22692 2. 23250 3. 24921 4. 25043 5. 25130 6. 25509 7. 16606 8. 16958 9. 17154 10. 17465 [ 178 others ]
3(1.1%)
3(1.1%)
3(1.1%)
3(1.1%)
3(1.1%)
3(1.1%)
2(0.8%)
2(0.8%)
2(0.8%)
2(0.8%)
238(90.1%)
 0 (0%)
2 S_4 [numeric] Mean (sd) : 17.6 (7.7) min < med < max: 5 < 24 < 24 IQR (CV) : 12 (0.4)
5:46(17.4%)
12:67(25.4%)
24:151(57.2%)
 0 (0%)
3 S_5 [numeric] AgeInDays Mean (sd) : 560.5 (249.6) min < med < max: 105 < 737.5 < 1097 IQR (CV) : 383.2 (0.4) 149 distinct values 0 (0%)
4 sacc_n_sce [numeric] E1SaccNum (the number of saccades) Mean (sd) : 82 (15.6) min < med < max: 38 < 81.5 < 170 IQR (CV) : 17.2 (0.2) 62 distinct values 0 (0%)
5 p1_sce [numeric] E1param1 (the value of a in v = b*d^a, with v = max displacement per sample in pixels and d = distance in pixels) Mean (sd) : 0.7 (0.1) min < med < max: 0.6 < 0.7 < 0.9 IQR (CV) : 0.1 (0.1) 262 distinct values 0 (0%)
6 p2_sce [numeric] E1param2 (the value of b in v = b*d^a) Mean (sd) : 0.7 (0.2) min < med < max: 0.2 < 0.7 < 1.4 IQR (CV) : 0.2 (0.3) 264 distinct values 0 (0%)
7 fix_n_sce [numeric] E1MeanFixNum (the mean number of fixation over 5 seconds of presentation for fixation quality == 1) Mean (sd) : 9.1 (2.4) min < med < max: 2.5 < 9.3 < 14.5 IQR (CV) : 3.3 (0.3) 73 distinct values 0 (0%)
8 fix_dur_sce [numeric] E1MeanFixDur (the mean duration of one fixation for fixation quality == 1) Mean (sd) : 370 (60.1) min < med < max: 272 < 358 < 697 IQR (CV) : 77 (0.2) 151 distinct values 0 (0%)
9 map_sce [numeric] E1MeanMapFix (the mean of density map values at fixations points for fixation quality == 1 | 2) Mean (sd) : 0.1 (0) min < med < max: 0 < 0.1 < 0.1 IQR (CV) : 0 (0.2) 264 distinct values 0 (0%)
10 look_t_all_sce [numeric] E1MeanLTAll (the mean of LT per image for fixation quality == 1 | 2) Mean (sd) : 4141.8 (311.5) min < med < max: 2515.3 < 4208 < 4916 IQR (CV) : 315.3 (0.1) 238 distinct values 0 (0%)
11 exp_t_sce [character] E1 expriment hour_min_sec 1. 09:30:24 2. 10:22:28 3. 10:28:42 4. 11:19:06 5. 08:46:40 6. 09:01:36 7. 09:06:04 8. 09:07:22 9. 09:07:24 10. 09:09:24 [ 250 others ]
2(0.8%)
2(0.8%)
2(0.8%)
2(0.8%)
1(0.4%)
1(0.4%)
1(0.4%)
1(0.4%)
1(0.4%)
1(0.4%)
250(94.7%)
 0 (0%)
12 sacc_n_fac [numeric] E2SaccNum Mean (sd) : 97.8 (15.9) min < med < max: 46 < 99 < 148 IQR (CV) : 20.5 (0.2) 69 distinct values 9 (3.41%)
13 p1_fac [numeric] E2param1 Mean (sd) : 0.7 (0.1) min < med < max: 0.5 < 0.7 < 0.9 IQR (CV) : 0.1 (0.1) 254 distinct values 9 (3.41%)
14 p2_fac [numeric] E2param2 Mean (sd) : 0.8 (0.2) min < med < max: 0.3 < 0.8 < 1.6 IQR (CV) : 0.3 (0.3) 254 distinct values 9 (3.41%)
15 fix_n_fac [numeric] E2MeanFixNum Mean (sd) : 8.3 (2.3) min < med < max: 2 < 8.2 < 15.8 IQR (CV) : 2.9 (0.3) 72 distinct values 9 (3.41%)
16 fix_dur_fac [numeric] E2MeanFixDur Mean (sd) : 382.8 (66.8) min < med < max: 222 < 372 < 607 IQR (CV) : 80.5 (0.2) 149 distinct values 9 (3.41%)
17 map_fac [numeric] E2MeanMapFix Mean (sd) : 0.1 (0) min < med < max: 0 < 0.1 < 0.1 IQR (CV) : 0 (0.2) 254 distinct values 9 (3.41%)
18 look_t_all_fac [numeric] E2MeanLTAll Mean (sd) : 4008.2 (353.9) min < med < max: 2155 < 4081 < 4645.5 IQR (CV) : 334.5 (0.1) 237 distinct values 9 (3.41%)
19 novelty_fac [numeric] E2Nov (the reaction to novel face over the four test [New-Old]./[New+Old]) Mean (sd) : 0.1 (0.2) min < med < max: -0.5 < 0.1 < 0.5 IQR (CV) : 0.2 (2) 255 distinct values 9 (3.41%)
20 reac_t_fac [numeric] E2RT (reaction time at face onset) Mean (sd) : 420.2 (84.8) min < med < max: 248 < 414.3 < 844 IQR (CV) : 104.8 (0.2) 237 distinct values 9 (3.41%)
21 pct_eyes_fac [numeric] E2Eyes (percent time on eyes) Mean (sd) : 0.7 (0.2) min < med < max: 0.1 < 0.7 < 1 IQR (CV) : 0.2 (0.3) 255 distinct values 9 (3.41%)
22 pct_mouth_fac [numeric] E2Mouth (percent time on mouth) Mean (sd) : 0.1 (0.2) min < med < max: 0 < 0.1 < 0.8 IQR (CV) : 0.2 (1.1) 235 distinct values 9 (3.41%)
23 exp_t_fac [character] E2 expriment hour_min_sec 1. 09:08:50 2. 10:22:40 3. 11:23:24 4. 08:48:06 5. 09:03:02 6. 09:08:32 7. 09:10:46 8. 09:13:28 9. 09:13:36 10. 09:14:46 [ 242 others ]
2(0.8%)
2(0.8%)
2(0.8%)
1(0.4%)
1(0.4%)
1(0.4%)
1(0.4%)
1(0.4%)
1(0.4%)
1(0.4%)
242(94.9%)
 9 (3.41%)
24 reac_t_tar [numeric] E3RT Mean (sd) : 291.4 (54.1) min < med < max: 182 < 286.9 < 484 IQR (CV) : 70.5 (0.2) 214 distinct values 4 (1.52%)
25 dist_tar [numeric] E3Dist (mean length of the first saccade to target) Mean (sd) : 187 (28) min < med < max: 128.3 < 185.4 < 345 IQR (CV) : 28.1 (0.1) 260 distinct values 4 (1.52%)
26 exp_t_tar [character] E3 expriment hour_min_sec 1. 10:51:14 2. 08:48:38 3. 09:03:40 4. 09:09:04 5. 09:09:30 6. 09:09:40 7. 09:11:36 8. 09:13:59 9. 09:14:04 10. 09:14:18 [ 250 others ]
2(0.8%)
1(0.4%)
1(0.4%)
1(0.4%)
1(0.4%)
1(0.4%)
1(0.4%)
1(0.4%)
1(0.4%)
1(0.4%)
250(95.8%)
 3 (1.14%)
27 reac_t_pur [numeric] E4RT Mean (sd) : 330 (88.3) min < med < max: 96 < 313 < 750 IQR (CV) : 89 (0.3) 171 distinct values 16 (6.06%)
28 exp_t_pur [character] E4 expriment hour_min_sec 1. 09:29:26 2. 09:36:44 3. 10:40:00 4. 08:49:56 5. 09:04:58 6. 09:10:08 7. 09:10:50 8. 09:12:36 9. 09:15:08 10. 09:15:12 [ 238 others ]
2(0.8%)
2(0.8%)
2(0.8%)
1(0.4%)
1(0.4%)
1(0.4%)
1(0.4%)
1(0.4%)
1(0.4%)
1(0.4%)
238(94.8%)
 13 (4.92%)

3.2 Score vs age

  • There seems to be a strong age effect hence I stratify the following descriptive analyses by group age

3.3 Score vs experiment hour

Here we look at the relationship between scores and experiment time. If there is a relationship it could be non monotonic (depending on sleep, meal times, etc). We look at mean score per hour for each age groups.

N exp per (rounded) hours (used experimentation time for face task):

  • Experiments mainly during the morning
  • 1 individual at 8 AM removed for the subsequent graphs (too much weight visually)

Next we look at the percent deviation from the mean for mean scores at each hour.

3.3.1 Scene exploration

  • Most variability for number of fixations
  • Shorter fixation durations in afternoonacross all age cats
  • different patterns according to age

3.3.2 Face exploration

Without novelty and pct_mouth (which showed a lot of variation):

  • Fixation durations 7-8% shorter in the afternoon: 1-2PM all cats + 12PM 24months
  • Consequently: fixation numbers higher at those times
  • Reaction times slower at those times
  • More looking at eyes in the afternoon (and probably consequently highr map mean)

For novelty and pct_mouth:

  • No pattern for novelty
  • Linked to previous remark on map meap and pct eyes, less looking at mouth on the afternoon

3.3.3 Target and pursuit

  • For 12M and 24M slower reaction times in the afternoon

3.3.4 Conclusion

For some scores there seems to be a non linear relationship with time of experiment, this should be included in the model.

3.4 Correlation between scores

Here we look at how the scores are correlated. We will focus on scores that are supposed to measure the same trait accross different tasks (eg reaction time at task 2 and 3) and scores that are mathematically linked (eg number of saccades and number of fixations).

3.4.1 Absolute correlations between all scores

  • all scores that measure saccade/fixations are grouped
  • the saccade equation parameters param1 and param2 highly correlated
  • the task 2 qualitative indicators pct_eyes, pct_mouth and map_mean correlated
  • correlations increase with age
  • reaction times not correlated

David (cf notes meeting 11/03/2020)

  • param1 donne une estimation de la « vitesse de base » du sujet
  • param2 = facteur de mise a l’échelle en fonction de l’amplitude
  • pas forcément de choix évident, on va commencer avec le param1

=> drop p2_sce & p2_fac

Entre n fix, n sacc et dur moyenne fix pour qual == 1 (qui sont les 3 liées) : Garder absolumùent la durée moyenne de fixation car normalement, représente l’état attentionnel de l’enfant (tps mis a traiter l’info regardée). Un comportement exploratoire va se traduire par un gd nb de fiwx, plus courtes

=> drop sacc_n_sce, sacc_n_fac, fix_n_sce & fix_n_fac

Pct eyes et mouth : choisir les yeux parceque c’est la cible privilégéies de l’attention visuelle des visages statiques (pas pareil si c’est un visage parlant). Yeux utilisés sur les études du trouble du spectre autistique par ex. Indicateur a prioi plus sensible de la cignition sociale de l’enfanrt. Attention car trop peu regarder ou trop regarder les yeux peut etre le signe d’un comporteùment atypique (peut etre regarder la distance a la moyenne).

Pour la taxche 2 on va enlever le mean map car on a déjà une mesure de l’endroit de la fixation par les yeux mais qui est plus indicatrice de de la typicité du comprtement.

=> drop pct_mouth_fac & map_fac

Remaining scores for analysis:

  • fix_dur_sce
  • p1_sce
  • map_sce
  • pct_eyes_fac
  • fix_dur_fac
  • p1_fac
  • reac_time_fac
  • reac_time_pur
  • novelty_fac
  • dist_tar
  • look_t_sce
  • look_t_fac

3.4.2 Correlation between tasks

Correlations on same scores/different tasks:

rowname score cor5 cor12 cor24
fix_n_sce fix_n_fac 0.62 0.47 0.51
fix_dur_sce fix_dur_fac 0.56 0.48 0.73
sacc_n_sce sacc_n_fac 0.54 0.52 0.57
reac_t_tar reac_t_pur 0.50 0.32 0.56
reac_t_fac reac_t_pur 0.40 0.19 0.28
look_t_all_sce look_t_all_fac 0.34 0.43 0.38
reac_t_fac reac_t_tar 0.22 0.34 0.35
  • The saccades/fixation parameters rather correlated accross task
  • Map mean and reaction time not so much
  • The reaction times even less
  • Overall increase of correlation with age

3.5 Grouping tasks

Given the previous comments, and after excluding less interpretable variables p1_sce, p1_fac and dist_tar we limit ourselves to the following indicators:

  • fix_dur_sce
  • map_sce
  • map_fac
  • pct_eyes_fac
  • fix_dur_fac
  • reac_time_fac
  • novelty_fac
  • reac_time_tar
  • reac_time_pur

3.5.1 Number of fixations

Number of fixations for scene (fix_dur_sce) and face (fix_dur_fac) tasks are comparable and will be grouped.

3.5.2 Reaction times

Reaction times at target (reac_t_tar) and pursuit (reac_t_pur) tasks are comparable (time constrained) but reaction time at face task (reac_t_fac) is different as there is no time contraint. Unfortunately reac_t_tar has many missing (166) so it will have to be excluded.

We have already seen that reac_t_tar has too many missing to be kept.

reac_t_fac task not correlated to two others. Also distributions differ. Which confirms we cannot combine it with reac_t_pur.

As reac_t_pur and reac_t_tar are more highly correlated, I suggest we keep reac_t_pur. UNLESS WE DECIDE THERE ARE TOO MANY MISSING AT PURSUIT TASK (5M = 7/39, 12M = 3/67, 24M = 6/151)

We could also keep reac_t_fac seperately, but I believe we want to try to limit the number of indicators we are using.

Or we could combine z-scores?

3.5.3 Looking location

First we decided to analyse the percentage of times looking at eyes at the face task (pct_eyes_fac) on its own because it is a well know marker of autistic spectrum disorders.

Then we want to compare the mean map location at the scene (map_sce) and face (map_fac) tasks:

Distributions are similar and scores are slightly correlated, we can combine them.

3.5.4 Looking time

Finally we wanted to add an indicator of attention quality, for this we decided to look at the looking time for all fixations look_t_all_sce and look_t_all_fac

3.5.5 Final groups

  • Group 1: fix_dur_sce, fix_dur_fac
  • Group 2: reac_time_pur
  • Group 3: pct_eyes_fac
  • Group 4: map_sce, map_fac
  • Group 5: novelty_fac
  • Group 6: look_t_all_sce, look_t_all_fac

3.6 Conclusions

  • Scores are linked with age
  • Scores are linked with time of experiment
  • Some variables are correlated hence the final selection of variables
  • The Correlations/groupings get clearer with age

Final groups:

  • Group 1 - mean fixation duration: fix_dur_sce, fix_dur_fac
  • Group 2 - reaction time: reac_time_pur
  • Group 3 - eye exploration: pct_eyes_fac
  • Group 4 - looking location: map_sce, map_fac
  • Group 5 - reaction to novelty: novelty_fac
  • Group 6 - attention quality: look_t_all_sce, look_t_all_fac

4 Correlation with other scores

4.1 Correlation with CBCL scores

Top 5 correlations:

babylab_task cbcl_task cor
fix_n_sce somscore 0.21
look_t_all_sce slescore 0.21
reac_t_fac slescore 0.18
pct_mouth_fac somscore 0.16
sacc_n_sce somscore 0.16

Bottom 5 correlations:

babylab_task cbcl_task cor
look_t_all_fac extscore -0.19
reac_t_pur othscore -0.19
look_t_all_fac aggscore -0.20
novelty_fac extscore -0.21
novelty_fac attscore -0.28

4.2 Correlation with MCHAT/MAB scores

Top 5 correlations:

babylab_task mchat_task cor
reac_t_fac MABwordsentprodscore_y2 0.26
fix_dur_sce MABwordsentlengthscore_y2 0.20
reac_t_fac MABwordsentlengthscore_y2 0.19
look_t_all_sce MABwordsentlengthscore_y2 0.17
pct_mouth_fac MABwordsentlengthscore_y2 0.17

Bottom 5 correlations:

babylab_task mchat_task cor
map_fac MABwordsentlengthscore_y2 -0.15
look_t_all_sce MABwordgestscore_y1 -0.16
p2_fac MABwordsentunderstdscore_y1 -0.17
look_t_all_fac MABwordsentunderstdscore_y1 -0.19
pct_eyes_fac MABwordsentprodscore_y1 -0.21

5 Annex: Repeated measures

5.1 Correlations

Correlation for each score between each period (5 months, 12 months and 24 months):

5.2 Visualisation

5.3 Conclusions

Not sure yet what we can do with the repeated measures.